xref: /openbmc/qemu/hw/arm/virt.c (revision 795c40b8)
1 /*
2  * ARM mach-virt emulation
3  *
4  * Copyright (c) 2013 Linaro Limited
5  *
6  * This program is free software; you can redistribute it and/or modify it
7  * under the terms and conditions of the GNU General Public License,
8  * version 2 or later, as published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope it will be useful, but WITHOUT
11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13  * more details.
14  *
15  * You should have received a copy of the GNU General Public License along with
16  * this program.  If not, see <http://www.gnu.org/licenses/>.
17  *
18  * Emulate a virtual board which works by passing Linux all the information
19  * it needs about what devices are present via the device tree.
20  * There are some restrictions about what we can do here:
21  *  + we can only present devices whose Linux drivers will work based
22  *    purely on the device tree with no platform data at all
23  *  + we want to present a very stripped-down minimalist platform,
24  *    both because this reduces the security attack surface from the guest
25  *    and also because it reduces our exposure to being broken when
26  *    the kernel updates its device tree bindings and requires further
27  *    information in a device binding that we aren't providing.
28  * This is essentially the same approach kvmtool uses.
29  */
30 
31 #include "qemu/osdep.h"
32 #include "qapi/error.h"
33 #include "hw/sysbus.h"
34 #include "hw/arm/arm.h"
35 #include "hw/arm/primecell.h"
36 #include "hw/arm/virt.h"
37 #include "hw/devices.h"
38 #include "net/net.h"
39 #include "sysemu/block-backend.h"
40 #include "sysemu/device_tree.h"
41 #include "sysemu/numa.h"
42 #include "sysemu/sysemu.h"
43 #include "sysemu/kvm.h"
44 #include "hw/compat.h"
45 #include "hw/loader.h"
46 #include "exec/address-spaces.h"
47 #include "qemu/bitops.h"
48 #include "qemu/error-report.h"
49 #include "hw/pci-host/gpex.h"
50 #include "hw/arm/sysbus-fdt.h"
51 #include "hw/platform-bus.h"
52 #include "hw/arm/fdt.h"
53 #include "hw/intc/arm_gic.h"
54 #include "hw/intc/arm_gicv3_common.h"
55 #include "kvm_arm.h"
56 #include "hw/smbios/smbios.h"
57 #include "qapi/visitor.h"
58 #include "standard-headers/linux/input.h"
59 
60 #define DEFINE_VIRT_MACHINE_LATEST(major, minor, latest) \
61     static void virt_##major##_##minor##_class_init(ObjectClass *oc, \
62                                                     void *data) \
63     { \
64         MachineClass *mc = MACHINE_CLASS(oc); \
65         virt_machine_##major##_##minor##_options(mc); \
66         mc->desc = "QEMU " # major "." # minor " ARM Virtual Machine"; \
67         if (latest) { \
68             mc->alias = "virt"; \
69         } \
70     } \
71     static const TypeInfo machvirt_##major##_##minor##_info = { \
72         .name = MACHINE_TYPE_NAME("virt-" # major "." # minor), \
73         .parent = TYPE_VIRT_MACHINE, \
74         .instance_init = virt_##major##_##minor##_instance_init, \
75         .class_init = virt_##major##_##minor##_class_init, \
76     }; \
77     static void machvirt_machine_##major##_##minor##_init(void) \
78     { \
79         type_register_static(&machvirt_##major##_##minor##_info); \
80     } \
81     type_init(machvirt_machine_##major##_##minor##_init);
82 
83 #define DEFINE_VIRT_MACHINE_AS_LATEST(major, minor) \
84     DEFINE_VIRT_MACHINE_LATEST(major, minor, true)
85 #define DEFINE_VIRT_MACHINE(major, minor) \
86     DEFINE_VIRT_MACHINE_LATEST(major, minor, false)
87 
88 
89 /* Number of external interrupt lines to configure the GIC with */
90 #define NUM_IRQS 256
91 
92 #define PLATFORM_BUS_NUM_IRQS 64
93 
94 static ARMPlatformBusSystemParams platform_bus_params;
95 
96 /* RAM limit in GB. Since VIRT_MEM starts at the 1GB mark, this means
97  * RAM can go up to the 256GB mark, leaving 256GB of the physical
98  * address space unallocated and free for future use between 256G and 512G.
99  * If we need to provide more RAM to VMs in the future then we need to:
100  *  * allocate a second bank of RAM starting at 2TB and working up
101  *  * fix the DT and ACPI table generation code in QEMU to correctly
102  *    report two split lumps of RAM to the guest
103  *  * fix KVM in the host kernel to allow guests with >40 bit address spaces
104  * (We don't want to fill all the way up to 512GB with RAM because
105  * we might want it for non-RAM purposes later. Conversely it seems
106  * reasonable to assume that anybody configuring a VM with a quarter
107  * of a terabyte of RAM will be doing it on a host with more than a
108  * terabyte of physical address space.)
109  */
110 #define RAMLIMIT_GB 255
111 #define RAMLIMIT_BYTES (RAMLIMIT_GB * 1024ULL * 1024 * 1024)
112 
113 /* Addresses and sizes of our components.
114  * 0..128MB is space for a flash device so we can run bootrom code such as UEFI.
115  * 128MB..256MB is used for miscellaneous device I/O.
116  * 256MB..1GB is reserved for possible future PCI support (ie where the
117  * PCI memory window will go if we add a PCI host controller).
118  * 1GB and up is RAM (which may happily spill over into the
119  * high memory region beyond 4GB).
120  * This represents a compromise between how much RAM can be given to
121  * a 32 bit VM and leaving space for expansion and in particular for PCI.
122  * Note that devices should generally be placed at multiples of 0x10000,
123  * to accommodate guests using 64K pages.
124  */
125 static const MemMapEntry a15memmap[] = {
126     /* Space up to 0x8000000 is reserved for a boot ROM */
127     [VIRT_FLASH] =              {          0, 0x08000000 },
128     [VIRT_CPUPERIPHS] =         { 0x08000000, 0x00020000 },
129     /* GIC distributor and CPU interfaces sit inside the CPU peripheral space */
130     [VIRT_GIC_DIST] =           { 0x08000000, 0x00010000 },
131     [VIRT_GIC_CPU] =            { 0x08010000, 0x00010000 },
132     [VIRT_GIC_V2M] =            { 0x08020000, 0x00001000 },
133     /* The space in between here is reserved for GICv3 CPU/vCPU/HYP */
134     [VIRT_GIC_ITS] =            { 0x08080000, 0x00020000 },
135     /* This redistributor space allows up to 2*64kB*123 CPUs */
136     [VIRT_GIC_REDIST] =         { 0x080A0000, 0x00F60000 },
137     [VIRT_UART] =               { 0x09000000, 0x00001000 },
138     [VIRT_RTC] =                { 0x09010000, 0x00001000 },
139     [VIRT_FW_CFG] =             { 0x09020000, 0x00000018 },
140     [VIRT_GPIO] =               { 0x09030000, 0x00001000 },
141     [VIRT_SECURE_UART] =        { 0x09040000, 0x00001000 },
142     [VIRT_MMIO] =               { 0x0a000000, 0x00000200 },
143     /* ...repeating for a total of NUM_VIRTIO_TRANSPORTS, each of that size */
144     [VIRT_PLATFORM_BUS] =       { 0x0c000000, 0x02000000 },
145     [VIRT_SECURE_MEM] =         { 0x0e000000, 0x01000000 },
146     [VIRT_PCIE_MMIO] =          { 0x10000000, 0x2eff0000 },
147     [VIRT_PCIE_PIO] =           { 0x3eff0000, 0x00010000 },
148     [VIRT_PCIE_ECAM] =          { 0x3f000000, 0x01000000 },
149     [VIRT_MEM] =                { 0x40000000, RAMLIMIT_BYTES },
150     /* Second PCIe window, 512GB wide at the 512GB boundary */
151     [VIRT_PCIE_MMIO_HIGH] =   { 0x8000000000ULL, 0x8000000000ULL },
152 };
153 
154 static const int a15irqmap[] = {
155     [VIRT_UART] = 1,
156     [VIRT_RTC] = 2,
157     [VIRT_PCIE] = 3, /* ... to 6 */
158     [VIRT_GPIO] = 7,
159     [VIRT_SECURE_UART] = 8,
160     [VIRT_MMIO] = 16, /* ...to 16 + NUM_VIRTIO_TRANSPORTS - 1 */
161     [VIRT_GIC_V2M] = 48, /* ...to 48 + NUM_GICV2M_SPIS - 1 */
162     [VIRT_PLATFORM_BUS] = 112, /* ...to 112 + PLATFORM_BUS_NUM_IRQS -1 */
163 };
164 
165 static const char *valid_cpus[] = {
166     "cortex-a15",
167     "cortex-a53",
168     "cortex-a57",
169     "host",
170 };
171 
172 static bool cpuname_valid(const char *cpu)
173 {
174     int i;
175 
176     for (i = 0; i < ARRAY_SIZE(valid_cpus); i++) {
177         if (strcmp(cpu, valid_cpus[i]) == 0) {
178             return true;
179         }
180     }
181     return false;
182 }
183 
184 static void create_fdt(VirtMachineState *vms)
185 {
186     void *fdt = create_device_tree(&vms->fdt_size);
187 
188     if (!fdt) {
189         error_report("create_device_tree() failed");
190         exit(1);
191     }
192 
193     vms->fdt = fdt;
194 
195     /* Header */
196     qemu_fdt_setprop_string(fdt, "/", "compatible", "linux,dummy-virt");
197     qemu_fdt_setprop_cell(fdt, "/", "#address-cells", 0x2);
198     qemu_fdt_setprop_cell(fdt, "/", "#size-cells", 0x2);
199 
200     /*
201      * /chosen and /memory nodes must exist for load_dtb
202      * to fill in necessary properties later
203      */
204     qemu_fdt_add_subnode(fdt, "/chosen");
205     qemu_fdt_add_subnode(fdt, "/memory");
206     qemu_fdt_setprop_string(fdt, "/memory", "device_type", "memory");
207 
208     /* Clock node, for the benefit of the UART. The kernel device tree
209      * binding documentation claims the PL011 node clock properties are
210      * optional but in practice if you omit them the kernel refuses to
211      * probe for the device.
212      */
213     vms->clock_phandle = qemu_fdt_alloc_phandle(fdt);
214     qemu_fdt_add_subnode(fdt, "/apb-pclk");
215     qemu_fdt_setprop_string(fdt, "/apb-pclk", "compatible", "fixed-clock");
216     qemu_fdt_setprop_cell(fdt, "/apb-pclk", "#clock-cells", 0x0);
217     qemu_fdt_setprop_cell(fdt, "/apb-pclk", "clock-frequency", 24000000);
218     qemu_fdt_setprop_string(fdt, "/apb-pclk", "clock-output-names",
219                                 "clk24mhz");
220     qemu_fdt_setprop_cell(fdt, "/apb-pclk", "phandle", vms->clock_phandle);
221 
222 }
223 
224 static void fdt_add_psci_node(const VirtMachineState *vms)
225 {
226     uint32_t cpu_suspend_fn;
227     uint32_t cpu_off_fn;
228     uint32_t cpu_on_fn;
229     uint32_t migrate_fn;
230     void *fdt = vms->fdt;
231     ARMCPU *armcpu = ARM_CPU(qemu_get_cpu(0));
232     const char *psci_method;
233 
234     switch (vms->psci_conduit) {
235     case QEMU_PSCI_CONDUIT_DISABLED:
236         return;
237     case QEMU_PSCI_CONDUIT_HVC:
238         psci_method = "hvc";
239         break;
240     case QEMU_PSCI_CONDUIT_SMC:
241         psci_method = "smc";
242         break;
243     default:
244         g_assert_not_reached();
245     }
246 
247     qemu_fdt_add_subnode(fdt, "/psci");
248     if (armcpu->psci_version == 2) {
249         const char comp[] = "arm,psci-0.2\0arm,psci";
250         qemu_fdt_setprop(fdt, "/psci", "compatible", comp, sizeof(comp));
251 
252         cpu_off_fn = QEMU_PSCI_0_2_FN_CPU_OFF;
253         if (arm_feature(&armcpu->env, ARM_FEATURE_AARCH64)) {
254             cpu_suspend_fn = QEMU_PSCI_0_2_FN64_CPU_SUSPEND;
255             cpu_on_fn = QEMU_PSCI_0_2_FN64_CPU_ON;
256             migrate_fn = QEMU_PSCI_0_2_FN64_MIGRATE;
257         } else {
258             cpu_suspend_fn = QEMU_PSCI_0_2_FN_CPU_SUSPEND;
259             cpu_on_fn = QEMU_PSCI_0_2_FN_CPU_ON;
260             migrate_fn = QEMU_PSCI_0_2_FN_MIGRATE;
261         }
262     } else {
263         qemu_fdt_setprop_string(fdt, "/psci", "compatible", "arm,psci");
264 
265         cpu_suspend_fn = QEMU_PSCI_0_1_FN_CPU_SUSPEND;
266         cpu_off_fn = QEMU_PSCI_0_1_FN_CPU_OFF;
267         cpu_on_fn = QEMU_PSCI_0_1_FN_CPU_ON;
268         migrate_fn = QEMU_PSCI_0_1_FN_MIGRATE;
269     }
270 
271     /* We adopt the PSCI spec's nomenclature, and use 'conduit' to refer
272      * to the instruction that should be used to invoke PSCI functions.
273      * However, the device tree binding uses 'method' instead, so that is
274      * what we should use here.
275      */
276     qemu_fdt_setprop_string(fdt, "/psci", "method", psci_method);
277 
278     qemu_fdt_setprop_cell(fdt, "/psci", "cpu_suspend", cpu_suspend_fn);
279     qemu_fdt_setprop_cell(fdt, "/psci", "cpu_off", cpu_off_fn);
280     qemu_fdt_setprop_cell(fdt, "/psci", "cpu_on", cpu_on_fn);
281     qemu_fdt_setprop_cell(fdt, "/psci", "migrate", migrate_fn);
282 }
283 
284 static void fdt_add_timer_nodes(const VirtMachineState *vms)
285 {
286     /* On real hardware these interrupts are level-triggered.
287      * On KVM they were edge-triggered before host kernel version 4.4,
288      * and level-triggered afterwards.
289      * On emulated QEMU they are level-triggered.
290      *
291      * Getting the DTB info about them wrong is awkward for some
292      * guest kernels:
293      *  pre-4.8 ignore the DT and leave the interrupt configured
294      *   with whatever the GIC reset value (or the bootloader) left it at
295      *  4.8 before rc6 honour the incorrect data by programming it back
296      *   into the GIC, causing problems
297      *  4.8rc6 and later ignore the DT and always write "level triggered"
298      *   into the GIC
299      *
300      * For backwards-compatibility, virt-2.8 and earlier will continue
301      * to say these are edge-triggered, but later machines will report
302      * the correct information.
303      */
304     ARMCPU *armcpu;
305     VirtMachineClass *vmc = VIRT_MACHINE_GET_CLASS(vms);
306     uint32_t irqflags = GIC_FDT_IRQ_FLAGS_LEVEL_HI;
307 
308     if (vmc->claim_edge_triggered_timers) {
309         irqflags = GIC_FDT_IRQ_FLAGS_EDGE_LO_HI;
310     }
311 
312     if (vms->gic_version == 2) {
313         irqflags = deposit32(irqflags, GIC_FDT_IRQ_PPI_CPU_START,
314                              GIC_FDT_IRQ_PPI_CPU_WIDTH,
315                              (1 << vms->smp_cpus) - 1);
316     }
317 
318     qemu_fdt_add_subnode(vms->fdt, "/timer");
319 
320     armcpu = ARM_CPU(qemu_get_cpu(0));
321     if (arm_feature(&armcpu->env, ARM_FEATURE_V8)) {
322         const char compat[] = "arm,armv8-timer\0arm,armv7-timer";
323         qemu_fdt_setprop(vms->fdt, "/timer", "compatible",
324                          compat, sizeof(compat));
325     } else {
326         qemu_fdt_setprop_string(vms->fdt, "/timer", "compatible",
327                                 "arm,armv7-timer");
328     }
329     qemu_fdt_setprop(vms->fdt, "/timer", "always-on", NULL, 0);
330     qemu_fdt_setprop_cells(vms->fdt, "/timer", "interrupts",
331                        GIC_FDT_IRQ_TYPE_PPI, ARCH_TIMER_S_EL1_IRQ, irqflags,
332                        GIC_FDT_IRQ_TYPE_PPI, ARCH_TIMER_NS_EL1_IRQ, irqflags,
333                        GIC_FDT_IRQ_TYPE_PPI, ARCH_TIMER_VIRT_IRQ, irqflags,
334                        GIC_FDT_IRQ_TYPE_PPI, ARCH_TIMER_NS_EL2_IRQ, irqflags);
335 }
336 
337 static void fdt_add_cpu_nodes(const VirtMachineState *vms)
338 {
339     int cpu;
340     int addr_cells = 1;
341     const MachineState *ms = MACHINE(vms);
342 
343     /*
344      * From Documentation/devicetree/bindings/arm/cpus.txt
345      *  On ARM v8 64-bit systems value should be set to 2,
346      *  that corresponds to the MPIDR_EL1 register size.
347      *  If MPIDR_EL1[63:32] value is equal to 0 on all CPUs
348      *  in the system, #address-cells can be set to 1, since
349      *  MPIDR_EL1[63:32] bits are not used for CPUs
350      *  identification.
351      *
352      *  Here we actually don't know whether our system is 32- or 64-bit one.
353      *  The simplest way to go is to examine affinity IDs of all our CPUs. If
354      *  at least one of them has Aff3 populated, we set #address-cells to 2.
355      */
356     for (cpu = 0; cpu < vms->smp_cpus; cpu++) {
357         ARMCPU *armcpu = ARM_CPU(qemu_get_cpu(cpu));
358 
359         if (armcpu->mp_affinity & ARM_AFF3_MASK) {
360             addr_cells = 2;
361             break;
362         }
363     }
364 
365     qemu_fdt_add_subnode(vms->fdt, "/cpus");
366     qemu_fdt_setprop_cell(vms->fdt, "/cpus", "#address-cells", addr_cells);
367     qemu_fdt_setprop_cell(vms->fdt, "/cpus", "#size-cells", 0x0);
368 
369     for (cpu = vms->smp_cpus - 1; cpu >= 0; cpu--) {
370         char *nodename = g_strdup_printf("/cpus/cpu@%d", cpu);
371         ARMCPU *armcpu = ARM_CPU(qemu_get_cpu(cpu));
372         CPUState *cs = CPU(armcpu);
373 
374         qemu_fdt_add_subnode(vms->fdt, nodename);
375         qemu_fdt_setprop_string(vms->fdt, nodename, "device_type", "cpu");
376         qemu_fdt_setprop_string(vms->fdt, nodename, "compatible",
377                                     armcpu->dtb_compatible);
378 
379         if (vms->psci_conduit != QEMU_PSCI_CONDUIT_DISABLED
380             && vms->smp_cpus > 1) {
381             qemu_fdt_setprop_string(vms->fdt, nodename,
382                                         "enable-method", "psci");
383         }
384 
385         if (addr_cells == 2) {
386             qemu_fdt_setprop_u64(vms->fdt, nodename, "reg",
387                                  armcpu->mp_affinity);
388         } else {
389             qemu_fdt_setprop_cell(vms->fdt, nodename, "reg",
390                                   armcpu->mp_affinity);
391         }
392 
393         if (ms->possible_cpus->cpus[cs->cpu_index].props.has_node_id) {
394             qemu_fdt_setprop_cell(vms->fdt, nodename, "numa-node-id",
395                 ms->possible_cpus->cpus[cs->cpu_index].props.node_id);
396         }
397 
398         g_free(nodename);
399     }
400 }
401 
402 static void fdt_add_its_gic_node(VirtMachineState *vms)
403 {
404     vms->msi_phandle = qemu_fdt_alloc_phandle(vms->fdt);
405     qemu_fdt_add_subnode(vms->fdt, "/intc/its");
406     qemu_fdt_setprop_string(vms->fdt, "/intc/its", "compatible",
407                             "arm,gic-v3-its");
408     qemu_fdt_setprop(vms->fdt, "/intc/its", "msi-controller", NULL, 0);
409     qemu_fdt_setprop_sized_cells(vms->fdt, "/intc/its", "reg",
410                                  2, vms->memmap[VIRT_GIC_ITS].base,
411                                  2, vms->memmap[VIRT_GIC_ITS].size);
412     qemu_fdt_setprop_cell(vms->fdt, "/intc/its", "phandle", vms->msi_phandle);
413 }
414 
415 static void fdt_add_v2m_gic_node(VirtMachineState *vms)
416 {
417     vms->msi_phandle = qemu_fdt_alloc_phandle(vms->fdt);
418     qemu_fdt_add_subnode(vms->fdt, "/intc/v2m");
419     qemu_fdt_setprop_string(vms->fdt, "/intc/v2m", "compatible",
420                             "arm,gic-v2m-frame");
421     qemu_fdt_setprop(vms->fdt, "/intc/v2m", "msi-controller", NULL, 0);
422     qemu_fdt_setprop_sized_cells(vms->fdt, "/intc/v2m", "reg",
423                                  2, vms->memmap[VIRT_GIC_V2M].base,
424                                  2, vms->memmap[VIRT_GIC_V2M].size);
425     qemu_fdt_setprop_cell(vms->fdt, "/intc/v2m", "phandle", vms->msi_phandle);
426 }
427 
428 static void fdt_add_gic_node(VirtMachineState *vms)
429 {
430     vms->gic_phandle = qemu_fdt_alloc_phandle(vms->fdt);
431     qemu_fdt_setprop_cell(vms->fdt, "/", "interrupt-parent", vms->gic_phandle);
432 
433     qemu_fdt_add_subnode(vms->fdt, "/intc");
434     qemu_fdt_setprop_cell(vms->fdt, "/intc", "#interrupt-cells", 3);
435     qemu_fdt_setprop(vms->fdt, "/intc", "interrupt-controller", NULL, 0);
436     qemu_fdt_setprop_cell(vms->fdt, "/intc", "#address-cells", 0x2);
437     qemu_fdt_setprop_cell(vms->fdt, "/intc", "#size-cells", 0x2);
438     qemu_fdt_setprop(vms->fdt, "/intc", "ranges", NULL, 0);
439     if (vms->gic_version == 3) {
440         qemu_fdt_setprop_string(vms->fdt, "/intc", "compatible",
441                                 "arm,gic-v3");
442         qemu_fdt_setprop_sized_cells(vms->fdt, "/intc", "reg",
443                                      2, vms->memmap[VIRT_GIC_DIST].base,
444                                      2, vms->memmap[VIRT_GIC_DIST].size,
445                                      2, vms->memmap[VIRT_GIC_REDIST].base,
446                                      2, vms->memmap[VIRT_GIC_REDIST].size);
447         if (vms->virt) {
448             qemu_fdt_setprop_cells(vms->fdt, "/intc", "interrupts",
449                                    GIC_FDT_IRQ_TYPE_PPI, ARCH_GICV3_MAINT_IRQ,
450                                    GIC_FDT_IRQ_FLAGS_LEVEL_HI);
451         }
452     } else {
453         /* 'cortex-a15-gic' means 'GIC v2' */
454         qemu_fdt_setprop_string(vms->fdt, "/intc", "compatible",
455                                 "arm,cortex-a15-gic");
456         qemu_fdt_setprop_sized_cells(vms->fdt, "/intc", "reg",
457                                       2, vms->memmap[VIRT_GIC_DIST].base,
458                                       2, vms->memmap[VIRT_GIC_DIST].size,
459                                       2, vms->memmap[VIRT_GIC_CPU].base,
460                                       2, vms->memmap[VIRT_GIC_CPU].size);
461     }
462 
463     qemu_fdt_setprop_cell(vms->fdt, "/intc", "phandle", vms->gic_phandle);
464 }
465 
466 static void fdt_add_pmu_nodes(const VirtMachineState *vms)
467 {
468     CPUState *cpu;
469     ARMCPU *armcpu;
470     uint32_t irqflags = GIC_FDT_IRQ_FLAGS_LEVEL_HI;
471 
472     CPU_FOREACH(cpu) {
473         armcpu = ARM_CPU(cpu);
474         if (!arm_feature(&armcpu->env, ARM_FEATURE_PMU) ||
475             (kvm_enabled() && !kvm_arm_pmu_create(cpu, PPI(VIRTUAL_PMU_IRQ)))) {
476             return;
477         }
478     }
479 
480     if (vms->gic_version == 2) {
481         irqflags = deposit32(irqflags, GIC_FDT_IRQ_PPI_CPU_START,
482                              GIC_FDT_IRQ_PPI_CPU_WIDTH,
483                              (1 << vms->smp_cpus) - 1);
484     }
485 
486     armcpu = ARM_CPU(qemu_get_cpu(0));
487     qemu_fdt_add_subnode(vms->fdt, "/pmu");
488     if (arm_feature(&armcpu->env, ARM_FEATURE_V8)) {
489         const char compat[] = "arm,armv8-pmuv3";
490         qemu_fdt_setprop(vms->fdt, "/pmu", "compatible",
491                          compat, sizeof(compat));
492         qemu_fdt_setprop_cells(vms->fdt, "/pmu", "interrupts",
493                                GIC_FDT_IRQ_TYPE_PPI, VIRTUAL_PMU_IRQ, irqflags);
494     }
495 }
496 
497 static void create_its(VirtMachineState *vms, DeviceState *gicdev)
498 {
499     const char *itsclass = its_class_name();
500     DeviceState *dev;
501 
502     if (!itsclass) {
503         /* Do nothing if not supported */
504         return;
505     }
506 
507     dev = qdev_create(NULL, itsclass);
508 
509     object_property_set_link(OBJECT(dev), OBJECT(gicdev), "parent-gicv3",
510                              &error_abort);
511     qdev_init_nofail(dev);
512     sysbus_mmio_map(SYS_BUS_DEVICE(dev), 0, vms->memmap[VIRT_GIC_ITS].base);
513 
514     fdt_add_its_gic_node(vms);
515 }
516 
517 static void create_v2m(VirtMachineState *vms, qemu_irq *pic)
518 {
519     int i;
520     int irq = vms->irqmap[VIRT_GIC_V2M];
521     DeviceState *dev;
522 
523     dev = qdev_create(NULL, "arm-gicv2m");
524     sysbus_mmio_map(SYS_BUS_DEVICE(dev), 0, vms->memmap[VIRT_GIC_V2M].base);
525     qdev_prop_set_uint32(dev, "base-spi", irq);
526     qdev_prop_set_uint32(dev, "num-spi", NUM_GICV2M_SPIS);
527     qdev_init_nofail(dev);
528 
529     for (i = 0; i < NUM_GICV2M_SPIS; i++) {
530         sysbus_connect_irq(SYS_BUS_DEVICE(dev), i, pic[irq + i]);
531     }
532 
533     fdt_add_v2m_gic_node(vms);
534 }
535 
536 static void create_gic(VirtMachineState *vms, qemu_irq *pic)
537 {
538     /* We create a standalone GIC */
539     DeviceState *gicdev;
540     SysBusDevice *gicbusdev;
541     const char *gictype;
542     int type = vms->gic_version, i;
543 
544     gictype = (type == 3) ? gicv3_class_name() : gic_class_name();
545 
546     gicdev = qdev_create(NULL, gictype);
547     qdev_prop_set_uint32(gicdev, "revision", type);
548     qdev_prop_set_uint32(gicdev, "num-cpu", smp_cpus);
549     /* Note that the num-irq property counts both internal and external
550      * interrupts; there are always 32 of the former (mandated by GIC spec).
551      */
552     qdev_prop_set_uint32(gicdev, "num-irq", NUM_IRQS + 32);
553     if (!kvm_irqchip_in_kernel()) {
554         qdev_prop_set_bit(gicdev, "has-security-extensions", vms->secure);
555     }
556     qdev_init_nofail(gicdev);
557     gicbusdev = SYS_BUS_DEVICE(gicdev);
558     sysbus_mmio_map(gicbusdev, 0, vms->memmap[VIRT_GIC_DIST].base);
559     if (type == 3) {
560         sysbus_mmio_map(gicbusdev, 1, vms->memmap[VIRT_GIC_REDIST].base);
561     } else {
562         sysbus_mmio_map(gicbusdev, 1, vms->memmap[VIRT_GIC_CPU].base);
563     }
564 
565     /* Wire the outputs from each CPU's generic timer and the GICv3
566      * maintenance interrupt signal to the appropriate GIC PPI inputs,
567      * and the GIC's IRQ/FIQ/VIRQ/VFIQ interrupt outputs to the CPU's inputs.
568      */
569     for (i = 0; i < smp_cpus; i++) {
570         DeviceState *cpudev = DEVICE(qemu_get_cpu(i));
571         int ppibase = NUM_IRQS + i * GIC_INTERNAL + GIC_NR_SGIS;
572         int irq;
573         /* Mapping from the output timer irq lines from the CPU to the
574          * GIC PPI inputs we use for the virt board.
575          */
576         const int timer_irq[] = {
577             [GTIMER_PHYS] = ARCH_TIMER_NS_EL1_IRQ,
578             [GTIMER_VIRT] = ARCH_TIMER_VIRT_IRQ,
579             [GTIMER_HYP]  = ARCH_TIMER_NS_EL2_IRQ,
580             [GTIMER_SEC]  = ARCH_TIMER_S_EL1_IRQ,
581         };
582 
583         for (irq = 0; irq < ARRAY_SIZE(timer_irq); irq++) {
584             qdev_connect_gpio_out(cpudev, irq,
585                                   qdev_get_gpio_in(gicdev,
586                                                    ppibase + timer_irq[irq]));
587         }
588 
589         qdev_connect_gpio_out_named(cpudev, "gicv3-maintenance-interrupt", 0,
590                                     qdev_get_gpio_in(gicdev, ppibase
591                                                      + ARCH_GICV3_MAINT_IRQ));
592 
593         sysbus_connect_irq(gicbusdev, i, qdev_get_gpio_in(cpudev, ARM_CPU_IRQ));
594         sysbus_connect_irq(gicbusdev, i + smp_cpus,
595                            qdev_get_gpio_in(cpudev, ARM_CPU_FIQ));
596         sysbus_connect_irq(gicbusdev, i + 2 * smp_cpus,
597                            qdev_get_gpio_in(cpudev, ARM_CPU_VIRQ));
598         sysbus_connect_irq(gicbusdev, i + 3 * smp_cpus,
599                            qdev_get_gpio_in(cpudev, ARM_CPU_VFIQ));
600     }
601 
602     for (i = 0; i < NUM_IRQS; i++) {
603         pic[i] = qdev_get_gpio_in(gicdev, i);
604     }
605 
606     fdt_add_gic_node(vms);
607 
608     if (type == 3 && vms->its) {
609         create_its(vms, gicdev);
610     } else if (type == 2) {
611         create_v2m(vms, pic);
612     }
613 }
614 
615 static void create_uart(const VirtMachineState *vms, qemu_irq *pic, int uart,
616                         MemoryRegion *mem, Chardev *chr)
617 {
618     char *nodename;
619     hwaddr base = vms->memmap[uart].base;
620     hwaddr size = vms->memmap[uart].size;
621     int irq = vms->irqmap[uart];
622     const char compat[] = "arm,pl011\0arm,primecell";
623     const char clocknames[] = "uartclk\0apb_pclk";
624     DeviceState *dev = qdev_create(NULL, "pl011");
625     SysBusDevice *s = SYS_BUS_DEVICE(dev);
626 
627     qdev_prop_set_chr(dev, "chardev", chr);
628     qdev_init_nofail(dev);
629     memory_region_add_subregion(mem, base,
630                                 sysbus_mmio_get_region(s, 0));
631     sysbus_connect_irq(s, 0, pic[irq]);
632 
633     nodename = g_strdup_printf("/pl011@%" PRIx64, base);
634     qemu_fdt_add_subnode(vms->fdt, nodename);
635     /* Note that we can't use setprop_string because of the embedded NUL */
636     qemu_fdt_setprop(vms->fdt, nodename, "compatible",
637                          compat, sizeof(compat));
638     qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg",
639                                      2, base, 2, size);
640     qemu_fdt_setprop_cells(vms->fdt, nodename, "interrupts",
641                                GIC_FDT_IRQ_TYPE_SPI, irq,
642                                GIC_FDT_IRQ_FLAGS_LEVEL_HI);
643     qemu_fdt_setprop_cells(vms->fdt, nodename, "clocks",
644                                vms->clock_phandle, vms->clock_phandle);
645     qemu_fdt_setprop(vms->fdt, nodename, "clock-names",
646                          clocknames, sizeof(clocknames));
647 
648     if (uart == VIRT_UART) {
649         qemu_fdt_setprop_string(vms->fdt, "/chosen", "stdout-path", nodename);
650     } else {
651         /* Mark as not usable by the normal world */
652         qemu_fdt_setprop_string(vms->fdt, nodename, "status", "disabled");
653         qemu_fdt_setprop_string(vms->fdt, nodename, "secure-status", "okay");
654     }
655 
656     g_free(nodename);
657 }
658 
659 static void create_rtc(const VirtMachineState *vms, qemu_irq *pic)
660 {
661     char *nodename;
662     hwaddr base = vms->memmap[VIRT_RTC].base;
663     hwaddr size = vms->memmap[VIRT_RTC].size;
664     int irq = vms->irqmap[VIRT_RTC];
665     const char compat[] = "arm,pl031\0arm,primecell";
666 
667     sysbus_create_simple("pl031", base, pic[irq]);
668 
669     nodename = g_strdup_printf("/pl031@%" PRIx64, base);
670     qemu_fdt_add_subnode(vms->fdt, nodename);
671     qemu_fdt_setprop(vms->fdt, nodename, "compatible", compat, sizeof(compat));
672     qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg",
673                                  2, base, 2, size);
674     qemu_fdt_setprop_cells(vms->fdt, nodename, "interrupts",
675                            GIC_FDT_IRQ_TYPE_SPI, irq,
676                            GIC_FDT_IRQ_FLAGS_LEVEL_HI);
677     qemu_fdt_setprop_cell(vms->fdt, nodename, "clocks", vms->clock_phandle);
678     qemu_fdt_setprop_string(vms->fdt, nodename, "clock-names", "apb_pclk");
679     g_free(nodename);
680 }
681 
682 static DeviceState *gpio_key_dev;
683 static void virt_powerdown_req(Notifier *n, void *opaque)
684 {
685     /* use gpio Pin 3 for power button event */
686     qemu_set_irq(qdev_get_gpio_in(gpio_key_dev, 0), 1);
687 }
688 
689 static Notifier virt_system_powerdown_notifier = {
690     .notify = virt_powerdown_req
691 };
692 
693 static void create_gpio(const VirtMachineState *vms, qemu_irq *pic)
694 {
695     char *nodename;
696     DeviceState *pl061_dev;
697     hwaddr base = vms->memmap[VIRT_GPIO].base;
698     hwaddr size = vms->memmap[VIRT_GPIO].size;
699     int irq = vms->irqmap[VIRT_GPIO];
700     const char compat[] = "arm,pl061\0arm,primecell";
701 
702     pl061_dev = sysbus_create_simple("pl061", base, pic[irq]);
703 
704     uint32_t phandle = qemu_fdt_alloc_phandle(vms->fdt);
705     nodename = g_strdup_printf("/pl061@%" PRIx64, base);
706     qemu_fdt_add_subnode(vms->fdt, nodename);
707     qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg",
708                                  2, base, 2, size);
709     qemu_fdt_setprop(vms->fdt, nodename, "compatible", compat, sizeof(compat));
710     qemu_fdt_setprop_cell(vms->fdt, nodename, "#gpio-cells", 2);
711     qemu_fdt_setprop(vms->fdt, nodename, "gpio-controller", NULL, 0);
712     qemu_fdt_setprop_cells(vms->fdt, nodename, "interrupts",
713                            GIC_FDT_IRQ_TYPE_SPI, irq,
714                            GIC_FDT_IRQ_FLAGS_LEVEL_HI);
715     qemu_fdt_setprop_cell(vms->fdt, nodename, "clocks", vms->clock_phandle);
716     qemu_fdt_setprop_string(vms->fdt, nodename, "clock-names", "apb_pclk");
717     qemu_fdt_setprop_cell(vms->fdt, nodename, "phandle", phandle);
718 
719     gpio_key_dev = sysbus_create_simple("gpio-key", -1,
720                                         qdev_get_gpio_in(pl061_dev, 3));
721     qemu_fdt_add_subnode(vms->fdt, "/gpio-keys");
722     qemu_fdt_setprop_string(vms->fdt, "/gpio-keys", "compatible", "gpio-keys");
723     qemu_fdt_setprop_cell(vms->fdt, "/gpio-keys", "#size-cells", 0);
724     qemu_fdt_setprop_cell(vms->fdt, "/gpio-keys", "#address-cells", 1);
725 
726     qemu_fdt_add_subnode(vms->fdt, "/gpio-keys/poweroff");
727     qemu_fdt_setprop_string(vms->fdt, "/gpio-keys/poweroff",
728                             "label", "GPIO Key Poweroff");
729     qemu_fdt_setprop_cell(vms->fdt, "/gpio-keys/poweroff", "linux,code",
730                           KEY_POWER);
731     qemu_fdt_setprop_cells(vms->fdt, "/gpio-keys/poweroff",
732                            "gpios", phandle, 3, 0);
733 
734     /* connect powerdown request */
735     qemu_register_powerdown_notifier(&virt_system_powerdown_notifier);
736 
737     g_free(nodename);
738 }
739 
740 static void create_virtio_devices(const VirtMachineState *vms, qemu_irq *pic)
741 {
742     int i;
743     hwaddr size = vms->memmap[VIRT_MMIO].size;
744 
745     /* We create the transports in forwards order. Since qbus_realize()
746      * prepends (not appends) new child buses, the incrementing loop below will
747      * create a list of virtio-mmio buses with decreasing base addresses.
748      *
749      * When a -device option is processed from the command line,
750      * qbus_find_recursive() picks the next free virtio-mmio bus in forwards
751      * order. The upshot is that -device options in increasing command line
752      * order are mapped to virtio-mmio buses with decreasing base addresses.
753      *
754      * When this code was originally written, that arrangement ensured that the
755      * guest Linux kernel would give the lowest "name" (/dev/vda, eth0, etc) to
756      * the first -device on the command line. (The end-to-end order is a
757      * function of this loop, qbus_realize(), qbus_find_recursive(), and the
758      * guest kernel's name-to-address assignment strategy.)
759      *
760      * Meanwhile, the kernel's traversal seems to have been reversed; see eg.
761      * the message, if not necessarily the code, of commit 70161ff336.
762      * Therefore the loop now establishes the inverse of the original intent.
763      *
764      * Unfortunately, we can't counteract the kernel change by reversing the
765      * loop; it would break existing command lines.
766      *
767      * In any case, the kernel makes no guarantee about the stability of
768      * enumeration order of virtio devices (as demonstrated by it changing
769      * between kernel versions). For reliable and stable identification
770      * of disks users must use UUIDs or similar mechanisms.
771      */
772     for (i = 0; i < NUM_VIRTIO_TRANSPORTS; i++) {
773         int irq = vms->irqmap[VIRT_MMIO] + i;
774         hwaddr base = vms->memmap[VIRT_MMIO].base + i * size;
775 
776         sysbus_create_simple("virtio-mmio", base, pic[irq]);
777     }
778 
779     /* We add dtb nodes in reverse order so that they appear in the finished
780      * device tree lowest address first.
781      *
782      * Note that this mapping is independent of the loop above. The previous
783      * loop influences virtio device to virtio transport assignment, whereas
784      * this loop controls how virtio transports are laid out in the dtb.
785      */
786     for (i = NUM_VIRTIO_TRANSPORTS - 1; i >= 0; i--) {
787         char *nodename;
788         int irq = vms->irqmap[VIRT_MMIO] + i;
789         hwaddr base = vms->memmap[VIRT_MMIO].base + i * size;
790 
791         nodename = g_strdup_printf("/virtio_mmio@%" PRIx64, base);
792         qemu_fdt_add_subnode(vms->fdt, nodename);
793         qemu_fdt_setprop_string(vms->fdt, nodename,
794                                 "compatible", "virtio,mmio");
795         qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg",
796                                      2, base, 2, size);
797         qemu_fdt_setprop_cells(vms->fdt, nodename, "interrupts",
798                                GIC_FDT_IRQ_TYPE_SPI, irq,
799                                GIC_FDT_IRQ_FLAGS_EDGE_LO_HI);
800         qemu_fdt_setprop(vms->fdt, nodename, "dma-coherent", NULL, 0);
801         g_free(nodename);
802     }
803 }
804 
805 static void create_one_flash(const char *name, hwaddr flashbase,
806                              hwaddr flashsize, const char *file,
807                              MemoryRegion *sysmem)
808 {
809     /* Create and map a single flash device. We use the same
810      * parameters as the flash devices on the Versatile Express board.
811      */
812     DriveInfo *dinfo = drive_get_next(IF_PFLASH);
813     DeviceState *dev = qdev_create(NULL, "cfi.pflash01");
814     SysBusDevice *sbd = SYS_BUS_DEVICE(dev);
815     const uint64_t sectorlength = 256 * 1024;
816 
817     if (dinfo) {
818         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
819                             &error_abort);
820     }
821 
822     qdev_prop_set_uint32(dev, "num-blocks", flashsize / sectorlength);
823     qdev_prop_set_uint64(dev, "sector-length", sectorlength);
824     qdev_prop_set_uint8(dev, "width", 4);
825     qdev_prop_set_uint8(dev, "device-width", 2);
826     qdev_prop_set_bit(dev, "big-endian", false);
827     qdev_prop_set_uint16(dev, "id0", 0x89);
828     qdev_prop_set_uint16(dev, "id1", 0x18);
829     qdev_prop_set_uint16(dev, "id2", 0x00);
830     qdev_prop_set_uint16(dev, "id3", 0x00);
831     qdev_prop_set_string(dev, "name", name);
832     qdev_init_nofail(dev);
833 
834     memory_region_add_subregion(sysmem, flashbase,
835                                 sysbus_mmio_get_region(SYS_BUS_DEVICE(dev), 0));
836 
837     if (file) {
838         char *fn;
839         int image_size;
840 
841         if (drive_get(IF_PFLASH, 0, 0)) {
842             error_report("The contents of the first flash device may be "
843                          "specified with -bios or with -drive if=pflash... "
844                          "but you cannot use both options at once");
845             exit(1);
846         }
847         fn = qemu_find_file(QEMU_FILE_TYPE_BIOS, file);
848         if (!fn) {
849             error_report("Could not find ROM image '%s'", file);
850             exit(1);
851         }
852         image_size = load_image_mr(fn, sysbus_mmio_get_region(sbd, 0));
853         g_free(fn);
854         if (image_size < 0) {
855             error_report("Could not load ROM image '%s'", file);
856             exit(1);
857         }
858     }
859 }
860 
861 static void create_flash(const VirtMachineState *vms,
862                          MemoryRegion *sysmem,
863                          MemoryRegion *secure_sysmem)
864 {
865     /* Create two flash devices to fill the VIRT_FLASH space in the memmap.
866      * Any file passed via -bios goes in the first of these.
867      * sysmem is the system memory space. secure_sysmem is the secure view
868      * of the system, and the first flash device should be made visible only
869      * there. The second flash device is visible to both secure and nonsecure.
870      * If sysmem == secure_sysmem this means there is no separate Secure
871      * address space and both flash devices are generally visible.
872      */
873     hwaddr flashsize = vms->memmap[VIRT_FLASH].size / 2;
874     hwaddr flashbase = vms->memmap[VIRT_FLASH].base;
875     char *nodename;
876 
877     create_one_flash("virt.flash0", flashbase, flashsize,
878                      bios_name, secure_sysmem);
879     create_one_flash("virt.flash1", flashbase + flashsize, flashsize,
880                      NULL, sysmem);
881 
882     if (sysmem == secure_sysmem) {
883         /* Report both flash devices as a single node in the DT */
884         nodename = g_strdup_printf("/flash@%" PRIx64, flashbase);
885         qemu_fdt_add_subnode(vms->fdt, nodename);
886         qemu_fdt_setprop_string(vms->fdt, nodename, "compatible", "cfi-flash");
887         qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg",
888                                      2, flashbase, 2, flashsize,
889                                      2, flashbase + flashsize, 2, flashsize);
890         qemu_fdt_setprop_cell(vms->fdt, nodename, "bank-width", 4);
891         g_free(nodename);
892     } else {
893         /* Report the devices as separate nodes so we can mark one as
894          * only visible to the secure world.
895          */
896         nodename = g_strdup_printf("/secflash@%" PRIx64, flashbase);
897         qemu_fdt_add_subnode(vms->fdt, nodename);
898         qemu_fdt_setprop_string(vms->fdt, nodename, "compatible", "cfi-flash");
899         qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg",
900                                      2, flashbase, 2, flashsize);
901         qemu_fdt_setprop_cell(vms->fdt, nodename, "bank-width", 4);
902         qemu_fdt_setprop_string(vms->fdt, nodename, "status", "disabled");
903         qemu_fdt_setprop_string(vms->fdt, nodename, "secure-status", "okay");
904         g_free(nodename);
905 
906         nodename = g_strdup_printf("/flash@%" PRIx64, flashbase);
907         qemu_fdt_add_subnode(vms->fdt, nodename);
908         qemu_fdt_setprop_string(vms->fdt, nodename, "compatible", "cfi-flash");
909         qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg",
910                                      2, flashbase + flashsize, 2, flashsize);
911         qemu_fdt_setprop_cell(vms->fdt, nodename, "bank-width", 4);
912         g_free(nodename);
913     }
914 }
915 
916 static FWCfgState *create_fw_cfg(const VirtMachineState *vms, AddressSpace *as)
917 {
918     hwaddr base = vms->memmap[VIRT_FW_CFG].base;
919     hwaddr size = vms->memmap[VIRT_FW_CFG].size;
920     FWCfgState *fw_cfg;
921     char *nodename;
922 
923     fw_cfg = fw_cfg_init_mem_wide(base + 8, base, 8, base + 16, as);
924     fw_cfg_add_i16(fw_cfg, FW_CFG_NB_CPUS, (uint16_t)smp_cpus);
925 
926     nodename = g_strdup_printf("/fw-cfg@%" PRIx64, base);
927     qemu_fdt_add_subnode(vms->fdt, nodename);
928     qemu_fdt_setprop_string(vms->fdt, nodename,
929                             "compatible", "qemu,fw-cfg-mmio");
930     qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg",
931                                  2, base, 2, size);
932     qemu_fdt_setprop(vms->fdt, nodename, "dma-coherent", NULL, 0);
933     g_free(nodename);
934     return fw_cfg;
935 }
936 
937 static void create_pcie_irq_map(const VirtMachineState *vms,
938                                 uint32_t gic_phandle,
939                                 int first_irq, const char *nodename)
940 {
941     int devfn, pin;
942     uint32_t full_irq_map[4 * 4 * 10] = { 0 };
943     uint32_t *irq_map = full_irq_map;
944 
945     for (devfn = 0; devfn <= 0x18; devfn += 0x8) {
946         for (pin = 0; pin < 4; pin++) {
947             int irq_type = GIC_FDT_IRQ_TYPE_SPI;
948             int irq_nr = first_irq + ((pin + PCI_SLOT(devfn)) % PCI_NUM_PINS);
949             int irq_level = GIC_FDT_IRQ_FLAGS_LEVEL_HI;
950             int i;
951 
952             uint32_t map[] = {
953                 devfn << 8, 0, 0,                           /* devfn */
954                 pin + 1,                                    /* PCI pin */
955                 gic_phandle, 0, 0, irq_type, irq_nr, irq_level }; /* GIC irq */
956 
957             /* Convert map to big endian */
958             for (i = 0; i < 10; i++) {
959                 irq_map[i] = cpu_to_be32(map[i]);
960             }
961             irq_map += 10;
962         }
963     }
964 
965     qemu_fdt_setprop(vms->fdt, nodename, "interrupt-map",
966                      full_irq_map, sizeof(full_irq_map));
967 
968     qemu_fdt_setprop_cells(vms->fdt, nodename, "interrupt-map-mask",
969                            0x1800, 0, 0, /* devfn (PCI_SLOT(3)) */
970                            0x7           /* PCI irq */);
971 }
972 
973 static void create_pcie(const VirtMachineState *vms, qemu_irq *pic)
974 {
975     hwaddr base_mmio = vms->memmap[VIRT_PCIE_MMIO].base;
976     hwaddr size_mmio = vms->memmap[VIRT_PCIE_MMIO].size;
977     hwaddr base_mmio_high = vms->memmap[VIRT_PCIE_MMIO_HIGH].base;
978     hwaddr size_mmio_high = vms->memmap[VIRT_PCIE_MMIO_HIGH].size;
979     hwaddr base_pio = vms->memmap[VIRT_PCIE_PIO].base;
980     hwaddr size_pio = vms->memmap[VIRT_PCIE_PIO].size;
981     hwaddr base_ecam = vms->memmap[VIRT_PCIE_ECAM].base;
982     hwaddr size_ecam = vms->memmap[VIRT_PCIE_ECAM].size;
983     hwaddr base = base_mmio;
984     int nr_pcie_buses = size_ecam / PCIE_MMCFG_SIZE_MIN;
985     int irq = vms->irqmap[VIRT_PCIE];
986     MemoryRegion *mmio_alias;
987     MemoryRegion *mmio_reg;
988     MemoryRegion *ecam_alias;
989     MemoryRegion *ecam_reg;
990     DeviceState *dev;
991     char *nodename;
992     int i;
993     PCIHostState *pci;
994 
995     dev = qdev_create(NULL, TYPE_GPEX_HOST);
996     qdev_init_nofail(dev);
997 
998     /* Map only the first size_ecam bytes of ECAM space */
999     ecam_alias = g_new0(MemoryRegion, 1);
1000     ecam_reg = sysbus_mmio_get_region(SYS_BUS_DEVICE(dev), 0);
1001     memory_region_init_alias(ecam_alias, OBJECT(dev), "pcie-ecam",
1002                              ecam_reg, 0, size_ecam);
1003     memory_region_add_subregion(get_system_memory(), base_ecam, ecam_alias);
1004 
1005     /* Map the MMIO window into system address space so as to expose
1006      * the section of PCI MMIO space which starts at the same base address
1007      * (ie 1:1 mapping for that part of PCI MMIO space visible through
1008      * the window).
1009      */
1010     mmio_alias = g_new0(MemoryRegion, 1);
1011     mmio_reg = sysbus_mmio_get_region(SYS_BUS_DEVICE(dev), 1);
1012     memory_region_init_alias(mmio_alias, OBJECT(dev), "pcie-mmio",
1013                              mmio_reg, base_mmio, size_mmio);
1014     memory_region_add_subregion(get_system_memory(), base_mmio, mmio_alias);
1015 
1016     if (vms->highmem) {
1017         /* Map high MMIO space */
1018         MemoryRegion *high_mmio_alias = g_new0(MemoryRegion, 1);
1019 
1020         memory_region_init_alias(high_mmio_alias, OBJECT(dev), "pcie-mmio-high",
1021                                  mmio_reg, base_mmio_high, size_mmio_high);
1022         memory_region_add_subregion(get_system_memory(), base_mmio_high,
1023                                     high_mmio_alias);
1024     }
1025 
1026     /* Map IO port space */
1027     sysbus_mmio_map(SYS_BUS_DEVICE(dev), 2, base_pio);
1028 
1029     for (i = 0; i < GPEX_NUM_IRQS; i++) {
1030         sysbus_connect_irq(SYS_BUS_DEVICE(dev), i, pic[irq + i]);
1031     }
1032 
1033     pci = PCI_HOST_BRIDGE(dev);
1034     if (pci->bus) {
1035         for (i = 0; i < nb_nics; i++) {
1036             NICInfo *nd = &nd_table[i];
1037 
1038             if (!nd->model) {
1039                 nd->model = g_strdup("virtio");
1040             }
1041 
1042             pci_nic_init_nofail(nd, pci->bus, nd->model, NULL);
1043         }
1044     }
1045 
1046     nodename = g_strdup_printf("/pcie@%" PRIx64, base);
1047     qemu_fdt_add_subnode(vms->fdt, nodename);
1048     qemu_fdt_setprop_string(vms->fdt, nodename,
1049                             "compatible", "pci-host-ecam-generic");
1050     qemu_fdt_setprop_string(vms->fdt, nodename, "device_type", "pci");
1051     qemu_fdt_setprop_cell(vms->fdt, nodename, "#address-cells", 3);
1052     qemu_fdt_setprop_cell(vms->fdt, nodename, "#size-cells", 2);
1053     qemu_fdt_setprop_cells(vms->fdt, nodename, "bus-range", 0,
1054                            nr_pcie_buses - 1);
1055     qemu_fdt_setprop(vms->fdt, nodename, "dma-coherent", NULL, 0);
1056 
1057     if (vms->msi_phandle) {
1058         qemu_fdt_setprop_cells(vms->fdt, nodename, "msi-parent",
1059                                vms->msi_phandle);
1060     }
1061 
1062     qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg",
1063                                  2, base_ecam, 2, size_ecam);
1064 
1065     if (vms->highmem) {
1066         qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "ranges",
1067                                      1, FDT_PCI_RANGE_IOPORT, 2, 0,
1068                                      2, base_pio, 2, size_pio,
1069                                      1, FDT_PCI_RANGE_MMIO, 2, base_mmio,
1070                                      2, base_mmio, 2, size_mmio,
1071                                      1, FDT_PCI_RANGE_MMIO_64BIT,
1072                                      2, base_mmio_high,
1073                                      2, base_mmio_high, 2, size_mmio_high);
1074     } else {
1075         qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "ranges",
1076                                      1, FDT_PCI_RANGE_IOPORT, 2, 0,
1077                                      2, base_pio, 2, size_pio,
1078                                      1, FDT_PCI_RANGE_MMIO, 2, base_mmio,
1079                                      2, base_mmio, 2, size_mmio);
1080     }
1081 
1082     qemu_fdt_setprop_cell(vms->fdt, nodename, "#interrupt-cells", 1);
1083     create_pcie_irq_map(vms, vms->gic_phandle, irq, nodename);
1084 
1085     g_free(nodename);
1086 }
1087 
1088 static void create_platform_bus(VirtMachineState *vms, qemu_irq *pic)
1089 {
1090     DeviceState *dev;
1091     SysBusDevice *s;
1092     int i;
1093     ARMPlatformBusFDTParams *fdt_params = g_new(ARMPlatformBusFDTParams, 1);
1094     MemoryRegion *sysmem = get_system_memory();
1095 
1096     platform_bus_params.platform_bus_base = vms->memmap[VIRT_PLATFORM_BUS].base;
1097     platform_bus_params.platform_bus_size = vms->memmap[VIRT_PLATFORM_BUS].size;
1098     platform_bus_params.platform_bus_first_irq = vms->irqmap[VIRT_PLATFORM_BUS];
1099     platform_bus_params.platform_bus_num_irqs = PLATFORM_BUS_NUM_IRQS;
1100 
1101     fdt_params->system_params = &platform_bus_params;
1102     fdt_params->binfo = &vms->bootinfo;
1103     fdt_params->intc = "/intc";
1104     /*
1105      * register a machine init done notifier that creates the device tree
1106      * nodes of the platform bus and its children dynamic sysbus devices
1107      */
1108     arm_register_platform_bus_fdt_creator(fdt_params);
1109 
1110     dev = qdev_create(NULL, TYPE_PLATFORM_BUS_DEVICE);
1111     dev->id = TYPE_PLATFORM_BUS_DEVICE;
1112     qdev_prop_set_uint32(dev, "num_irqs",
1113         platform_bus_params.platform_bus_num_irqs);
1114     qdev_prop_set_uint32(dev, "mmio_size",
1115         platform_bus_params.platform_bus_size);
1116     qdev_init_nofail(dev);
1117     s = SYS_BUS_DEVICE(dev);
1118 
1119     for (i = 0; i < platform_bus_params.platform_bus_num_irqs; i++) {
1120         int irqn = platform_bus_params.platform_bus_first_irq + i;
1121         sysbus_connect_irq(s, i, pic[irqn]);
1122     }
1123 
1124     memory_region_add_subregion(sysmem,
1125                                 platform_bus_params.platform_bus_base,
1126                                 sysbus_mmio_get_region(s, 0));
1127 }
1128 
1129 static void create_secure_ram(VirtMachineState *vms,
1130                               MemoryRegion *secure_sysmem)
1131 {
1132     MemoryRegion *secram = g_new(MemoryRegion, 1);
1133     char *nodename;
1134     hwaddr base = vms->memmap[VIRT_SECURE_MEM].base;
1135     hwaddr size = vms->memmap[VIRT_SECURE_MEM].size;
1136 
1137     memory_region_init_ram(secram, NULL, "virt.secure-ram", size, &error_fatal);
1138     vmstate_register_ram_global(secram);
1139     memory_region_add_subregion(secure_sysmem, base, secram);
1140 
1141     nodename = g_strdup_printf("/secram@%" PRIx64, base);
1142     qemu_fdt_add_subnode(vms->fdt, nodename);
1143     qemu_fdt_setprop_string(vms->fdt, nodename, "device_type", "memory");
1144     qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg", 2, base, 2, size);
1145     qemu_fdt_setprop_string(vms->fdt, nodename, "status", "disabled");
1146     qemu_fdt_setprop_string(vms->fdt, nodename, "secure-status", "okay");
1147 
1148     g_free(nodename);
1149 }
1150 
1151 static void *machvirt_dtb(const struct arm_boot_info *binfo, int *fdt_size)
1152 {
1153     const VirtMachineState *board = container_of(binfo, VirtMachineState,
1154                                                  bootinfo);
1155 
1156     *fdt_size = board->fdt_size;
1157     return board->fdt;
1158 }
1159 
1160 static void virt_build_smbios(VirtMachineState *vms)
1161 {
1162     uint8_t *smbios_tables, *smbios_anchor;
1163     size_t smbios_tables_len, smbios_anchor_len;
1164     const char *product = "QEMU Virtual Machine";
1165 
1166     if (!vms->fw_cfg) {
1167         return;
1168     }
1169 
1170     if (kvm_enabled()) {
1171         product = "KVM Virtual Machine";
1172     }
1173 
1174     smbios_set_defaults("QEMU", product,
1175                         "1.0", false, true, SMBIOS_ENTRY_POINT_30);
1176 
1177     smbios_get_tables(NULL, 0, &smbios_tables, &smbios_tables_len,
1178                       &smbios_anchor, &smbios_anchor_len);
1179 
1180     if (smbios_anchor) {
1181         fw_cfg_add_file(vms->fw_cfg, "etc/smbios/smbios-tables",
1182                         smbios_tables, smbios_tables_len);
1183         fw_cfg_add_file(vms->fw_cfg, "etc/smbios/smbios-anchor",
1184                         smbios_anchor, smbios_anchor_len);
1185     }
1186 }
1187 
1188 static
1189 void virt_machine_done(Notifier *notifier, void *data)
1190 {
1191     VirtMachineState *vms = container_of(notifier, VirtMachineState,
1192                                          machine_done);
1193 
1194     virt_acpi_setup(vms);
1195     virt_build_smbios(vms);
1196 }
1197 
1198 static uint64_t virt_cpu_mp_affinity(VirtMachineState *vms, int idx)
1199 {
1200     uint8_t clustersz = ARM_DEFAULT_CPUS_PER_CLUSTER;
1201     VirtMachineClass *vmc = VIRT_MACHINE_GET_CLASS(vms);
1202 
1203     if (!vmc->disallow_affinity_adjustment) {
1204         /* Adjust MPIDR like 64-bit KVM hosts, which incorporate the
1205          * GIC's target-list limitations. 32-bit KVM hosts currently
1206          * always create clusters of 4 CPUs, but that is expected to
1207          * change when they gain support for gicv3. When KVM is enabled
1208          * it will override the changes we make here, therefore our
1209          * purposes are to make TCG consistent (with 64-bit KVM hosts)
1210          * and to improve SGI efficiency.
1211          */
1212         if (vms->gic_version == 3) {
1213             clustersz = GICV3_TARGETLIST_BITS;
1214         } else {
1215             clustersz = GIC_TARGETLIST_BITS;
1216         }
1217     }
1218     return arm_cpu_mp_affinity(idx, clustersz);
1219 }
1220 
1221 static void machvirt_init(MachineState *machine)
1222 {
1223     VirtMachineState *vms = VIRT_MACHINE(machine);
1224     VirtMachineClass *vmc = VIRT_MACHINE_GET_CLASS(machine);
1225     MachineClass *mc = MACHINE_GET_CLASS(machine);
1226     const CPUArchIdList *possible_cpus;
1227     qemu_irq pic[NUM_IRQS];
1228     MemoryRegion *sysmem = get_system_memory();
1229     MemoryRegion *secure_sysmem = NULL;
1230     int n, virt_max_cpus;
1231     MemoryRegion *ram = g_new(MemoryRegion, 1);
1232     const char *cpu_model = machine->cpu_model;
1233     char **cpustr;
1234     ObjectClass *oc;
1235     const char *typename;
1236     CPUClass *cc;
1237     Error *err = NULL;
1238     bool firmware_loaded = bios_name || drive_get(IF_PFLASH, 0, 0);
1239 
1240     if (!cpu_model) {
1241         cpu_model = "cortex-a15";
1242     }
1243 
1244     /* We can probe only here because during property set
1245      * KVM is not available yet
1246      */
1247     if (!vms->gic_version) {
1248         if (!kvm_enabled()) {
1249             error_report("gic-version=host requires KVM");
1250             exit(1);
1251         }
1252 
1253         vms->gic_version = kvm_arm_vgic_probe();
1254         if (!vms->gic_version) {
1255             error_report("Unable to determine GIC version supported by host");
1256             exit(1);
1257         }
1258     }
1259 
1260     /* Separate the actual CPU model name from any appended features */
1261     cpustr = g_strsplit(cpu_model, ",", 2);
1262 
1263     if (!cpuname_valid(cpustr[0])) {
1264         error_report("mach-virt: CPU %s not supported", cpustr[0]);
1265         exit(1);
1266     }
1267 
1268     /* If we have an EL3 boot ROM then the assumption is that it will
1269      * implement PSCI itself, so disable QEMU's internal implementation
1270      * so it doesn't get in the way. Instead of starting secondary
1271      * CPUs in PSCI powerdown state we will start them all running and
1272      * let the boot ROM sort them out.
1273      * The usual case is that we do use QEMU's PSCI implementation;
1274      * if the guest has EL2 then we will use SMC as the conduit,
1275      * and otherwise we will use HVC (for backwards compatibility and
1276      * because if we're using KVM then we must use HVC).
1277      */
1278     if (vms->secure && firmware_loaded) {
1279         vms->psci_conduit = QEMU_PSCI_CONDUIT_DISABLED;
1280     } else if (vms->virt) {
1281         vms->psci_conduit = QEMU_PSCI_CONDUIT_SMC;
1282     } else {
1283         vms->psci_conduit = QEMU_PSCI_CONDUIT_HVC;
1284     }
1285 
1286     /* The maximum number of CPUs depends on the GIC version, or on how
1287      * many redistributors we can fit into the memory map.
1288      */
1289     if (vms->gic_version == 3) {
1290         virt_max_cpus = vms->memmap[VIRT_GIC_REDIST].size / 0x20000;
1291     } else {
1292         virt_max_cpus = GIC_NCPU;
1293     }
1294 
1295     if (max_cpus > virt_max_cpus) {
1296         error_report("Number of SMP CPUs requested (%d) exceeds max CPUs "
1297                      "supported by machine 'mach-virt' (%d)",
1298                      max_cpus, virt_max_cpus);
1299         exit(1);
1300     }
1301 
1302     vms->smp_cpus = smp_cpus;
1303 
1304     if (machine->ram_size > vms->memmap[VIRT_MEM].size) {
1305         error_report("mach-virt: cannot model more than %dGB RAM", RAMLIMIT_GB);
1306         exit(1);
1307     }
1308 
1309     if (vms->virt && kvm_enabled()) {
1310         error_report("mach-virt: KVM does not support providing "
1311                      "Virtualization extensions to the guest CPU");
1312         exit(1);
1313     }
1314 
1315     if (vms->secure) {
1316         if (kvm_enabled()) {
1317             error_report("mach-virt: KVM does not support Security extensions");
1318             exit(1);
1319         }
1320 
1321         /* The Secure view of the world is the same as the NonSecure,
1322          * but with a few extra devices. Create it as a container region
1323          * containing the system memory at low priority; any secure-only
1324          * devices go in at higher priority and take precedence.
1325          */
1326         secure_sysmem = g_new(MemoryRegion, 1);
1327         memory_region_init(secure_sysmem, OBJECT(machine), "secure-memory",
1328                            UINT64_MAX);
1329         memory_region_add_subregion_overlap(secure_sysmem, 0, sysmem, -1);
1330     }
1331 
1332     create_fdt(vms);
1333 
1334     oc = cpu_class_by_name(TYPE_ARM_CPU, cpustr[0]);
1335     if (!oc) {
1336         error_report("Unable to find CPU definition");
1337         exit(1);
1338     }
1339     typename = object_class_get_name(oc);
1340 
1341     /* convert -smp CPU options specified by the user into global props */
1342     cc = CPU_CLASS(oc);
1343     cc->parse_features(typename, cpustr[1], &err);
1344     g_strfreev(cpustr);
1345     if (err) {
1346         error_report_err(err);
1347         exit(1);
1348     }
1349 
1350     possible_cpus = mc->possible_cpu_arch_ids(machine);
1351     for (n = 0; n < possible_cpus->len; n++) {
1352         Object *cpuobj;
1353         CPUState *cs;
1354         int node_id;
1355 
1356         if (n >= smp_cpus) {
1357             break;
1358         }
1359 
1360         cpuobj = object_new(typename);
1361         object_property_set_int(cpuobj, possible_cpus->cpus[n].arch_id,
1362                                 "mp-affinity", NULL);
1363 
1364         cs = CPU(cpuobj);
1365         cs->cpu_index = n;
1366 
1367         node_id = possible_cpus->cpus[cs->cpu_index].props.node_id;
1368         if (!possible_cpus->cpus[cs->cpu_index].props.has_node_id) {
1369             /* by default CPUState::numa_node was 0 if it's not set via CLI
1370              * keep it this way for now but in future we probably should
1371              * refuse to start up with incomplete numa mapping */
1372              node_id = 0;
1373         }
1374         if (cs->numa_node == CPU_UNSET_NUMA_NODE_ID) {
1375             cs->numa_node = node_id;
1376         } else {
1377             /* CPU isn't device_add compatible yet, this shouldn't happen */
1378             error_setg(&error_abort, "user set node-id not implemented");
1379         }
1380 
1381         if (!vms->secure) {
1382             object_property_set_bool(cpuobj, false, "has_el3", NULL);
1383         }
1384 
1385         if (!vms->virt && object_property_find(cpuobj, "has_el2", NULL)) {
1386             object_property_set_bool(cpuobj, false, "has_el2", NULL);
1387         }
1388 
1389         if (vms->psci_conduit != QEMU_PSCI_CONDUIT_DISABLED) {
1390             object_property_set_int(cpuobj, vms->psci_conduit,
1391                                     "psci-conduit", NULL);
1392 
1393             /* Secondary CPUs start in PSCI powered-down state */
1394             if (n > 0) {
1395                 object_property_set_bool(cpuobj, true,
1396                                          "start-powered-off", NULL);
1397             }
1398         }
1399 
1400         if (vmc->no_pmu && object_property_find(cpuobj, "pmu", NULL)) {
1401             object_property_set_bool(cpuobj, false, "pmu", NULL);
1402         }
1403 
1404         if (object_property_find(cpuobj, "reset-cbar", NULL)) {
1405             object_property_set_int(cpuobj, vms->memmap[VIRT_CPUPERIPHS].base,
1406                                     "reset-cbar", &error_abort);
1407         }
1408 
1409         object_property_set_link(cpuobj, OBJECT(sysmem), "memory",
1410                                  &error_abort);
1411         if (vms->secure) {
1412             object_property_set_link(cpuobj, OBJECT(secure_sysmem),
1413                                      "secure-memory", &error_abort);
1414         }
1415 
1416         object_property_set_bool(cpuobj, true, "realized", NULL);
1417         object_unref(cpuobj);
1418     }
1419     fdt_add_timer_nodes(vms);
1420     fdt_add_cpu_nodes(vms);
1421     fdt_add_psci_node(vms);
1422 
1423     memory_region_allocate_system_memory(ram, NULL, "mach-virt.ram",
1424                                          machine->ram_size);
1425     memory_region_add_subregion(sysmem, vms->memmap[VIRT_MEM].base, ram);
1426 
1427     create_flash(vms, sysmem, secure_sysmem ? secure_sysmem : sysmem);
1428 
1429     create_gic(vms, pic);
1430 
1431     fdt_add_pmu_nodes(vms);
1432 
1433     create_uart(vms, pic, VIRT_UART, sysmem, serial_hds[0]);
1434 
1435     if (vms->secure) {
1436         create_secure_ram(vms, secure_sysmem);
1437         create_uart(vms, pic, VIRT_SECURE_UART, secure_sysmem, serial_hds[1]);
1438     }
1439 
1440     create_rtc(vms, pic);
1441 
1442     create_pcie(vms, pic);
1443 
1444     create_gpio(vms, pic);
1445 
1446     /* Create mmio transports, so the user can create virtio backends
1447      * (which will be automatically plugged in to the transports). If
1448      * no backend is created the transport will just sit harmlessly idle.
1449      */
1450     create_virtio_devices(vms, pic);
1451 
1452     vms->fw_cfg = create_fw_cfg(vms, &address_space_memory);
1453     rom_set_fw(vms->fw_cfg);
1454 
1455     vms->machine_done.notify = virt_machine_done;
1456     qemu_add_machine_init_done_notifier(&vms->machine_done);
1457 
1458     vms->bootinfo.ram_size = machine->ram_size;
1459     vms->bootinfo.kernel_filename = machine->kernel_filename;
1460     vms->bootinfo.kernel_cmdline = machine->kernel_cmdline;
1461     vms->bootinfo.initrd_filename = machine->initrd_filename;
1462     vms->bootinfo.nb_cpus = smp_cpus;
1463     vms->bootinfo.board_id = -1;
1464     vms->bootinfo.loader_start = vms->memmap[VIRT_MEM].base;
1465     vms->bootinfo.get_dtb = machvirt_dtb;
1466     vms->bootinfo.firmware_loaded = firmware_loaded;
1467     arm_load_kernel(ARM_CPU(first_cpu), &vms->bootinfo);
1468 
1469     /*
1470      * arm_load_kernel machine init done notifier registration must
1471      * happen before the platform_bus_create call. In this latter,
1472      * another notifier is registered which adds platform bus nodes.
1473      * Notifiers are executed in registration reverse order.
1474      */
1475     create_platform_bus(vms, pic);
1476 }
1477 
1478 static bool virt_get_secure(Object *obj, Error **errp)
1479 {
1480     VirtMachineState *vms = VIRT_MACHINE(obj);
1481 
1482     return vms->secure;
1483 }
1484 
1485 static void virt_set_secure(Object *obj, bool value, Error **errp)
1486 {
1487     VirtMachineState *vms = VIRT_MACHINE(obj);
1488 
1489     vms->secure = value;
1490 }
1491 
1492 static bool virt_get_virt(Object *obj, Error **errp)
1493 {
1494     VirtMachineState *vms = VIRT_MACHINE(obj);
1495 
1496     return vms->virt;
1497 }
1498 
1499 static void virt_set_virt(Object *obj, bool value, Error **errp)
1500 {
1501     VirtMachineState *vms = VIRT_MACHINE(obj);
1502 
1503     vms->virt = value;
1504 }
1505 
1506 static bool virt_get_highmem(Object *obj, Error **errp)
1507 {
1508     VirtMachineState *vms = VIRT_MACHINE(obj);
1509 
1510     return vms->highmem;
1511 }
1512 
1513 static void virt_set_highmem(Object *obj, bool value, Error **errp)
1514 {
1515     VirtMachineState *vms = VIRT_MACHINE(obj);
1516 
1517     vms->highmem = value;
1518 }
1519 
1520 static bool virt_get_its(Object *obj, Error **errp)
1521 {
1522     VirtMachineState *vms = VIRT_MACHINE(obj);
1523 
1524     return vms->its;
1525 }
1526 
1527 static void virt_set_its(Object *obj, bool value, Error **errp)
1528 {
1529     VirtMachineState *vms = VIRT_MACHINE(obj);
1530 
1531     vms->its = value;
1532 }
1533 
1534 static char *virt_get_gic_version(Object *obj, Error **errp)
1535 {
1536     VirtMachineState *vms = VIRT_MACHINE(obj);
1537     const char *val = vms->gic_version == 3 ? "3" : "2";
1538 
1539     return g_strdup(val);
1540 }
1541 
1542 static void virt_set_gic_version(Object *obj, const char *value, Error **errp)
1543 {
1544     VirtMachineState *vms = VIRT_MACHINE(obj);
1545 
1546     if (!strcmp(value, "3")) {
1547         vms->gic_version = 3;
1548     } else if (!strcmp(value, "2")) {
1549         vms->gic_version = 2;
1550     } else if (!strcmp(value, "host")) {
1551         vms->gic_version = 0; /* Will probe later */
1552     } else {
1553         error_setg(errp, "Invalid gic-version value");
1554         error_append_hint(errp, "Valid values are 3, 2, host.\n");
1555     }
1556 }
1557 
1558 static CpuInstanceProperties
1559 virt_cpu_index_to_props(MachineState *ms, unsigned cpu_index)
1560 {
1561     MachineClass *mc = MACHINE_GET_CLASS(ms);
1562     const CPUArchIdList *possible_cpus = mc->possible_cpu_arch_ids(ms);
1563 
1564     assert(cpu_index < possible_cpus->len);
1565     return possible_cpus->cpus[cpu_index].props;
1566 }
1567 
1568 static const CPUArchIdList *virt_possible_cpu_arch_ids(MachineState *ms)
1569 {
1570     int n;
1571     VirtMachineState *vms = VIRT_MACHINE(ms);
1572 
1573     if (ms->possible_cpus) {
1574         assert(ms->possible_cpus->len == max_cpus);
1575         return ms->possible_cpus;
1576     }
1577 
1578     ms->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
1579                                   sizeof(CPUArchId) * max_cpus);
1580     ms->possible_cpus->len = max_cpus;
1581     for (n = 0; n < ms->possible_cpus->len; n++) {
1582         ms->possible_cpus->cpus[n].arch_id =
1583             virt_cpu_mp_affinity(vms, n);
1584         ms->possible_cpus->cpus[n].props.has_thread_id = true;
1585         ms->possible_cpus->cpus[n].props.thread_id = n;
1586 
1587         /* default distribution of CPUs over NUMA nodes */
1588         if (nb_numa_nodes) {
1589             /* preset values but do not enable them i.e. 'has_node_id = false',
1590              * numa init code will enable them later if manual mapping wasn't
1591              * present on CLI */
1592             ms->possible_cpus->cpus[n].props.node_id = n % nb_numa_nodes;
1593         }
1594     }
1595     return ms->possible_cpus;
1596 }
1597 
1598 static void virt_machine_class_init(ObjectClass *oc, void *data)
1599 {
1600     MachineClass *mc = MACHINE_CLASS(oc);
1601 
1602     mc->init = machvirt_init;
1603     /* Start max_cpus at the maximum QEMU supports. We'll further restrict
1604      * it later in machvirt_init, where we have more information about the
1605      * configuration of the particular instance.
1606      */
1607     mc->max_cpus = 255;
1608     mc->has_dynamic_sysbus = true;
1609     mc->block_default_type = IF_VIRTIO;
1610     mc->no_cdrom = 1;
1611     mc->pci_allow_0_address = true;
1612     /* We know we will never create a pre-ARMv7 CPU which needs 1K pages */
1613     mc->minimum_page_bits = 12;
1614     mc->possible_cpu_arch_ids = virt_possible_cpu_arch_ids;
1615     mc->cpu_index_to_instance_props = virt_cpu_index_to_props;
1616 }
1617 
1618 static const TypeInfo virt_machine_info = {
1619     .name          = TYPE_VIRT_MACHINE,
1620     .parent        = TYPE_MACHINE,
1621     .abstract      = true,
1622     .instance_size = sizeof(VirtMachineState),
1623     .class_size    = sizeof(VirtMachineClass),
1624     .class_init    = virt_machine_class_init,
1625 };
1626 
1627 static void machvirt_machine_init(void)
1628 {
1629     type_register_static(&virt_machine_info);
1630 }
1631 type_init(machvirt_machine_init);
1632 
1633 static void virt_2_9_instance_init(Object *obj)
1634 {
1635     VirtMachineState *vms = VIRT_MACHINE(obj);
1636     VirtMachineClass *vmc = VIRT_MACHINE_GET_CLASS(vms);
1637 
1638     /* EL3 is disabled by default on virt: this makes us consistent
1639      * between KVM and TCG for this board, and it also allows us to
1640      * boot UEFI blobs which assume no TrustZone support.
1641      */
1642     vms->secure = false;
1643     object_property_add_bool(obj, "secure", virt_get_secure,
1644                              virt_set_secure, NULL);
1645     object_property_set_description(obj, "secure",
1646                                     "Set on/off to enable/disable the ARM "
1647                                     "Security Extensions (TrustZone)",
1648                                     NULL);
1649 
1650     /* EL2 is also disabled by default, for similar reasons */
1651     vms->virt = false;
1652     object_property_add_bool(obj, "virtualization", virt_get_virt,
1653                              virt_set_virt, NULL);
1654     object_property_set_description(obj, "virtualization",
1655                                     "Set on/off to enable/disable emulating a "
1656                                     "guest CPU which implements the ARM "
1657                                     "Virtualization Extensions",
1658                                     NULL);
1659 
1660     /* High memory is enabled by default */
1661     vms->highmem = true;
1662     object_property_add_bool(obj, "highmem", virt_get_highmem,
1663                              virt_set_highmem, NULL);
1664     object_property_set_description(obj, "highmem",
1665                                     "Set on/off to enable/disable using "
1666                                     "physical address space above 32 bits",
1667                                     NULL);
1668     /* Default GIC type is v2 */
1669     vms->gic_version = 2;
1670     object_property_add_str(obj, "gic-version", virt_get_gic_version,
1671                         virt_set_gic_version, NULL);
1672     object_property_set_description(obj, "gic-version",
1673                                     "Set GIC version. "
1674                                     "Valid values are 2, 3 and host", NULL);
1675 
1676     if (vmc->no_its) {
1677         vms->its = false;
1678     } else {
1679         /* Default allows ITS instantiation */
1680         vms->its = true;
1681         object_property_add_bool(obj, "its", virt_get_its,
1682                                  virt_set_its, NULL);
1683         object_property_set_description(obj, "its",
1684                                         "Set on/off to enable/disable "
1685                                         "ITS instantiation",
1686                                         NULL);
1687     }
1688 
1689     vms->memmap = a15memmap;
1690     vms->irqmap = a15irqmap;
1691 }
1692 
1693 static void virt_machine_2_9_options(MachineClass *mc)
1694 {
1695 }
1696 DEFINE_VIRT_MACHINE_AS_LATEST(2, 9)
1697 
1698 #define VIRT_COMPAT_2_8 \
1699     HW_COMPAT_2_8
1700 
1701 static void virt_2_8_instance_init(Object *obj)
1702 {
1703     virt_2_9_instance_init(obj);
1704 }
1705 
1706 static void virt_machine_2_8_options(MachineClass *mc)
1707 {
1708     VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
1709 
1710     virt_machine_2_9_options(mc);
1711     SET_MACHINE_COMPAT(mc, VIRT_COMPAT_2_8);
1712     /* For 2.8 and earlier we falsely claimed in the DT that
1713      * our timers were edge-triggered, not level-triggered.
1714      */
1715     vmc->claim_edge_triggered_timers = true;
1716 }
1717 DEFINE_VIRT_MACHINE(2, 8)
1718 
1719 #define VIRT_COMPAT_2_7 \
1720     HW_COMPAT_2_7
1721 
1722 static void virt_2_7_instance_init(Object *obj)
1723 {
1724     virt_2_8_instance_init(obj);
1725 }
1726 
1727 static void virt_machine_2_7_options(MachineClass *mc)
1728 {
1729     VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
1730 
1731     virt_machine_2_8_options(mc);
1732     SET_MACHINE_COMPAT(mc, VIRT_COMPAT_2_7);
1733     /* ITS was introduced with 2.8 */
1734     vmc->no_its = true;
1735     /* Stick with 1K pages for migration compatibility */
1736     mc->minimum_page_bits = 0;
1737 }
1738 DEFINE_VIRT_MACHINE(2, 7)
1739 
1740 #define VIRT_COMPAT_2_6 \
1741     HW_COMPAT_2_6
1742 
1743 static void virt_2_6_instance_init(Object *obj)
1744 {
1745     virt_2_7_instance_init(obj);
1746 }
1747 
1748 static void virt_machine_2_6_options(MachineClass *mc)
1749 {
1750     VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
1751 
1752     virt_machine_2_7_options(mc);
1753     SET_MACHINE_COMPAT(mc, VIRT_COMPAT_2_6);
1754     vmc->disallow_affinity_adjustment = true;
1755     /* Disable PMU for 2.6 as PMU support was first introduced in 2.7 */
1756     vmc->no_pmu = true;
1757 }
1758 DEFINE_VIRT_MACHINE(2, 6)
1759